作者: Bradley C. Love , Marc Tomlinson , Todd M. Gureckis
DOI: 10.1016/S0079-7421(08)00005-4
关键词: Generalization (learning) 、 Reinforcement learning 、 Computer science 、 Point (typography) 、 Cognitive science 、 Action (philosophy) 、 Range (mathematics) 、 Opposition (planets) 、 Mental representation 、 Stimulus (psychology)
摘要: We live in a world consisting of concrete experiences, yet we appear to form abstractions that transcend the details our experiences. In this contribution, argue abstract nature thought is overstated and representations are inherently bound examples experience during learning. present three lines related research support general point. The first line suggests there no separate learning systems for acquiring mental rules storing exceptions these rules. Instead, both items types share common representational substrate grounded experienced training examples. second concepts, such as same different can range over an unbounded set stimulus properties, rooted coupled with analogical processes. Finally, consider how people perform dynamic decision tasks which short‐ long‐term rewards opposition. Rather than invoking explicit reasoning processes planning, people's performance best explained by reinforcement procedures update estimates action values reactive, trial‐by‐trial fashion. All implicate mechanisms capable broad generalization, local terms used updating planning future actions. end considering benefits designing operate according principles.